A Method Of Ship Moving Target Tracking And Trajectory Acquisition Based On Multi-features

With the popularity of surveillance cameras,more and more information is collected through the camera.How to extract effective information from these video has gradually become a hot research issue in the field of computer vision.The research on the tracking and trajectory acquisition methods of the ship based on multi-features in this paper is based on this direction.For tracking and trajectory acquisition of ship,the strategy adopted in this paper is to identify the post-match tracking first.For the purpose of identification at moving ship target,in order to realize the recognition of ship targets in complex scenarios,the YOLO algorithm that has made breakthrough progress in end-to-end real-time detection will be prototyped at this article,and the characteristics of the ship are extracted by manual self-learning features of the machine are fused,and the original algorithm is improved to enable video-based moving ship target recognition.The artificially extracted HOG features and LBP features were combined using the weighted combination feature fusion method.The fusion features were then re-integrated with the neural network self-extracted features.Through actual experiments,the results show that using the video with simple background environments and good resolution as input data,the method of this paper can basically identify the ship targets in the image.For the purpose of how to achieve the matching of target moving slices between frames,this paper uses an improved SIFT matching algorithm.Since the traditional SIFT description operators are generally high-dimensional data and does not conducive to data calculation,the dimensionality reduction of the data is achieved in this paper through the MDS dimensionality reduction algorithm.Simultaneously with the HOG feature,the feature mismatch matching point is eliminated twice,so that the matching points are simplified and the area is precisely matched,thereby further reducing the adverse effects of occlusion and ship-to-ship coordination.Through practical experiments,the results show that the improved matching algorithm can further reduce false matching points and eliminate matching points generated in the background area.Through the matching method,the association establishment of the same ship slice between different frames can be realized to achieve the purpose of tracking.Comparing the tracking method of this paper with TLD algorithm,the method of this paper does not need manual intervention when tracking.Compared the tracking method of this paper with the improved Kalman filter tracking method,the method of this paper is more accurate in tracking.During the tracking process,the method of this paper extracts the coordinates of each slice and stores this information in the specified file.The name of the specified file is determined according to the slice number.In this way,the same numbered slice coordinate data is stored in the same file,so as to obtain the trajectory data of the ship's target,and the trajectory data is visualized in the next frame of video.